Hi.
I created a basic completion prompt to get the AI to assess a student’s short written email. For some reason, it keeps getting things wrong.
My initial prompt was very long, with examples and templates and detailed instructions, but that was giving me inaccurate outcomes. So I thought I’d distill it down to something very basic to see if that made it more accurate, and it still keeps outputting inaccurate responses.
Here’s my simplified prompt:
You are an expert in English language assessment. Your students are intermediate level EFL (English as a Foreign Language) students.
*The following is an email written by a student: *
###
Hi Sue.
I am writing this email to tell you that i want you to study with in my house.
*Come to my house on friday evening. *
I think the english exam is easy but we need to study hard.
see you soon!
Abdulaziz
###
Identify the errors in grammar and punctuation in the student’s writing:
1. “I am writing this email to tell you that i want you” - Missing a verb in the independent clause (e.g. “I am writing this email to tell you that I want you to study with me in my house”).
2. “study with in my house” - Wrong preposition (“with” instead of “in”).
3. “Come to my house on friday evening” - Missing article (“Come to the my house…”).
4. “I think the english exam is easy” - Missing capitalization for “English”.
5. “but we need to study hard” - Missing comma after “easy”.
6. “See you soon!” - Missing exclamation mark.
I then adjusted the temperature to the opposite end and got the following response:
1. “I am writing this email to tell you that i want you” - should be “I am writing this email to tell you that I want you”
2. “Come to my house on friday evening” - should be “Come to my house on Friday evening”
3. “I think the english exam is easy” - should be “I think the English exam is easy”
4. “but we need to study hard” - should be “but we need to study hard.” (add a period at the end)
5. “see you soon!” - should be “See you soon!” (capitalize the first letter)
In both cases, the inaccuracy rate is too high for it to be a meaningful output.
Is there something I’m doing wrong in my prompt? Is there a way to get accurate responses? Has anyone managed it and is happy to share how?